Like this presentation? Why not share!

9,128

-1

-1

Published on

Cluster Analysis

No Downloads

Total Views

9,128

On Slideshare

0

From Embeds

0

Number of Embeds

1

Shares

0

Downloads

0

Comments

1

Likes

13

No embeds

No notes for slide

- 1. Cluster Analysis: Basic Concepts and Algorithms<br />
- 2. What is Cluster Analysis?<br />Finding groups of objects such that the objects in a group will be similar (or related) to one another and different from (or unrelated to) the objects in other groups<br />
- 3. Applications of Cluster Analysis <br />Understanding<br />Group genes and proteins that have similar functionality, or group stocks with similar price fluctuations<br />Summarization<br />Reduce the size of large data sets <br />
- 4. Types of Clustering <br />A clustering is a set of clusters <br /> Important distinction between hierarchical and partitional sets of clusters <br /> Partitional Clustering <br /> A division data objects into non-overlapping subsets (clusters) such that each data object is in exactly one subset <br /> Hierarchical clustering <br /> A set of nested clusters organized as a hierarchical tree <br />
- 5. Clustering Algorithms<br />K-means <br />Hierarchical clustering <br />Graph based clustering <br />
- 6. K-means Clustering <br />Partitional clustering approach <br />Each cluster is associated with a centroid (center point) <br />Each point is assigned to the cluster with the closest centroid<br />Number of clusters, K, must be specified <br />The basic algorithm is very simple <br />
- 7. K-means Clustering – Details <br />Initial centroids are often chosen randomly. <br />Clusters produced vary from one run to another. <br />The centroid is (typically) the mean of the points in the cluster. <br />‘Closeness’ is measured by Euclidean distance, cosine similarity, correlation, etc. <br />K-means will converge for common similarity measures mentioned above<br />
- 8. K-means Clustering – Details <br />Most of the convergence happens in the first few iterations. <br />Often the stopping condition is changed to ‘Until relatively few points change clusters’ <br />Complexity is O( n * K * I * d ) <br />n = number of points, K = number of clusters, I = number of iterations, d = number of attributes <br />
- 9. Two different K-means Clusterings <br />Sub-optimal Clustering <br />Optimal Clustering <br />
- 10. Problems with Selecting Initial Points <br />If there are K ‘real’ clusters then the chance of selecting one centroid from each cluster is small. <br />Chance is relatively small when K is large <br />If clusters are the same size, n, then For example, if K = 10, then probability = 10!/1010 = 0.00036 <br />Sometimes the initial centroids will readjust themselves in ‘right’ way, and sometimes they don’t <br />Consider an example of five pairs of clusters <br />
- 11. Solutions to Initial Centroids Problem<br />Multiple runs <br />Helps, but probability is not on your side <br /> Sample and use hierarchical clustering to determine initial centroids<br /> Select more than k initial centroids and then select among these initial centroids<br />Select most widely separated <br /> Bisecting K-means <br />Not as susceptible to initialization issues <br />
- 12. Evaluating K-means Clusters <br />Most common measure is Sum of Squared Error (SSE) <br />For each point, the error is the distance to the nearest cluster <br />To get SSE, we square these errors and sum them. <br /> x is a data point in cluster Ciand mi is the representative point for cluster Ci<br />can show that micorresponds to the center (mean) of the cluster <br />
- 13. Evaluating K-means Clusters <br />Given two clusters, we can choose the one with the smaller error <br />One easy way to reduce SSE is to increase K, the number of clusters <br />A good clustering with smaller K can have a lower SSE than a poor clustering with higher K <br />
- 14. Limitations of K-means <br />K-means has problems when clusters are of differing <br />Sizes <br />Densities <br />Non-globular shapes <br /> K-means has problems when the data contains outliers. <br /> The number of clusters (K) is difficult to determine. <br />
- 15. Hierarchical Clustering <br />Produces a set of nested clusters organized as a hierarchical tree <br />Can be visualized as a dendrogram<br />A tree like diagram that records the sequences of merges or splits <br />
- 16. Strengths of Hierarchical Clustering <br />Do not have to assume any particular number of clusters <br />Any desired number of clusters can be obtained by ‘cutting’ the dendogram at the proper level <br /> They may correspond to meaningful taxonomies <br />Example in biological sciences (e.g., animal kingdom, phylogeny reconstruction, …) <br />
- 17. Hierarchical Clustering <br />Two main types of hierarchical clustering <br />Agglomerative: <br />Start with the points as individual clusters <br />At each step, merge the closest pair of clusters until only one cluster (or k clusters) left <br />Divisive: <br />Start with one, all-inclusive cluster <br />At each step, split a cluster until each cluster contains a point (or there are k clusters) <br />
- 18. Agglomerative Clustering Algorithm <br />More popular hierarchical clustering technique <br />Basic algorithm is straightforward <br />Compute the proximity matrix <br />Let each data point be a cluster<br />Repeat <br /> Merge the two closest clusters <br /> Update the proximity matrix <br />Until only a single cluster remains <br />
- 19. Hierarchical Clustering: Group Average <br />Compromise between Single and Complete Link <br /> Strengths <br />Less susceptible to noise and outliers <br /> Limitations <br />Biased towards globular clusters <br />
- 20. Hierarchical Clustering: Time and Space requirements <br />O(N2) space since it uses the proximity matrix. <br />N is the number of points. <br /> O(N3) time in many cases <br />There are N steps and at each step the size, N2, proximity matrix must be updated and searched <br />Complexity can be reduced to O(N2 log(N) ) time for some approaches <br />
- 21. Hierarchical Clustering: Problems and Limitations <br />Once a decision is made to combine two clusters, it cannot be undone No objective function is directly minimized <br />Different schemes have problems with one or more of the following: <br />Sensitivity to noise and outliers (MIN) <br />Difficulty handling different sized clusters and non-convex shapes (Group average, MAX) <br />Breaking large clusters (MAX) <br />
- 22. conclusion<br />The purpose of clustering in data mining and its types are discussed.<br />The k-means and hierarchical algorithm are explained in detail and their pros and cons are analyzed.<br />
- 23. Visit more self help tutorials<br />Pick a tutorial of your choice and browse through it at your own pace.<br />The tutorials section is free, self-guiding and will not involve any additional support.<br />Visit us at www.dataminingtools.net<br />

No public clipboards found for this slide

×
### Save the most important slides with Clipping

Clipping is a handy way to collect and organize the most important slides from a presentation. You can keep your great finds in clipboards organized around topics.

But I can not load the document, I would like to help send the Email puangkaew_Msit04@gmail.com and CC to boy1999@hotmail.com.

Thank you very much.